Applied Business Forecasting and

Planning

MOVING AVERAGES AND

EXPONENTIAL SMOOTHING
Introduction
This chapter introduces models applicable to time series
data with seasonal, trend, or both seasonal and trend
component and stationary data.
Forecasting methods discussed in this chapter can be
classified as:
Averaging methods.
Equally weighted observations
Exponential Smoothing methods.
Unequal set of weights to past data, where the weights decay
exponentially from the most recent to the most distant data points.
All methods in this group require that certain parameters to be
defined.
These parameters (with values between 0 and 1) will determine the
unequal weights to be applied to past data.
Introduction
Averaging methods
If a time series is generated by a constant process
subject to random error, then mean is a useful statistic
and can be used as a forecast for the next period.
Averaging methods are suitable for stationary time
series data where the series is in equilibrium around a
constant value ( the underlying mean) with a constant
variance over time.
Introduction
Exponential smoothing methods
The simplest exponential smoothing method is the
single smoothing (SES) method where only one
parameter needs to be estimated
Holts method makes use of two different parameters
and allows forecasting for series with trend.
Holt-Winters method involves three smoothing
parameters to smooth the data, the trend, and the
seasonal index.
Averaging Methods
The Mean
Uses the average of all the historical data as the forecast
1 t
Ft 1 yi
t i 1
When new data becomes available , the forecast for
time t+2 is the new mean including the previously
observed data plus this new observation.
1 t 1
Ft 2
t 1 i 1
yi

This method is appropriate when there is no noticeable

trend or seasonality.
Averaging Methods
The moving average for time period t is the
mean of the k most recent observations.
The constant number k is specified at the
outset.
The smaller the number k, the more weight
is given to recent periods.
The greater the number k, the less weight is
given to more recent periods.
Moving Averages
A large k is desirable when there are wide,
infrequent fluctuations in the series.
A small k is most desirable when there are
sudden shifts in the level of series.
For quarterly data, a four-quarter moving
average, MA(4), eliminates or averages out
seasonal effects.
Moving Averages
For monthly data, a 12-month moving
average, MA(12), eliminate or averages out
seasonal effect.
Equal weights are assigned to each
observation used in the average.
Each new data point is included in the
average as it becomes available, and the
oldest data point is discarded.
Moving Averages
A moving average of order k, MA(k) is the value
of k consecutive observations.
( yt yt 1 yt 2 yt k 1 )
Ft 1 y t 1
K
1 t
Ft 1 yi
k i t k 1
K is the number of terms in the moving average.
The moving average model does not handle trend
or seasonality very well although it can do better
than the total mean.
Example: Weekly Department Store Sales
Period (t) Sales (y)

The weekly sales 1

2
5.3
4.4
3 5.4

figures (in millions of 4

5
5.8
5.6

dollars) presented in 6
7
8
4.8
5.6
5.6

the following table are 9

10
5.4
6.5
11 5.1

used by a major 12
13
5.8
5

department store to 14
15
16
6.2
5.6
6.7

determine the need for 17

18
5.2
5.5
19 5.8

temporary sales 20
21
5.1
5.8

personnel. 22
23
24
6.7
5.2
6
25 5.8
 Example: Weekly Department Store Sales
Weekly Sales

5
Sales

4 Sales (y)

0
0 5 10 15 20 25 30

Weeks
Example: Weekly Department Store Sales

Use a three-week moving average (k=3) for

the department store sales to forecast for the
week 24 and 26.
( y23 y22 y21 ) 5.2 6.7 5.8
y 24 5.9
3 3
The forecast error is
e24 y24 y 24 6 5.9 .1
Example: Weekly Department Store Sales

The forecast for the week 26 is

y25 y24 y23 5.8 6 5.2
y 26 5.7
3 3
 Example: Weekly Department Store Sales
Period (t) Sales (y) forecast
RMSE = 0.63 1 5.3
2 4.4
3 5.4
Weekly Sales Forecasts
4 5.8 5.033333
5 5.6 5.2
8
6 4.8 5.6
7 5.6 5.4
7
8 5.6 5.333333
9 5.4 5.333333
6 10 6.5 5.533333
11 5.1 5.833333
5 12 5.8 5.666667
13 5 5.8
Sales

Sales (y)
4 14 6.2 5.3
forecast
15 5.6 5.666667
16 6.7 5.6
3
17 5.2 6.166667
18 5.5 5.833333
2
19 5.8 5.8
20 5.1 5.5
1
21 5.8 5.466667
22 6.7 5.566667
0 23 5.2 5.866667
0 5 10 15 20 25 30
24 6 5.9
Weeks
25 5.8 5.966667
5.666667
Exponential Smoothing Methods
This method provides an exponentially
weighted moving average of all previously
observed values.
Appropriate for data with no predictable
upward or downward trend.
The aim is to estimate the current level and
use it as a forecast of future value.
Simple Exponential Smoothing Method
Formally, the exponential smoothing equation is
Ft 1 yt (1 ) Ft
Ft 1 forecast for the next period.
= smoothing constant.
yt = observed value of series in period t.
Ft = old forecast for period t.
The forecast Ft+1 is based on weighting the most recent
observation yt with a weight and weighting the most
recent forecast Ft with a weight of 1-
Simple Exponential Smoothing Method
The implication of exponential smoothing
can be better seen if the previous equation is
expanded by replacing Ft with its
components as follows:
Ft 1 yt (1 ) Ft
yt (1 )[ yt 1 (1 ) Ft 1 ]
yt (1 ) y t 1 (1 ) 2 Ft 1
 Simple Exponential Smoothing Method
If this substitution process is repeated by
replacing Ft-1 by its components, Ft-2 by its
components, and so on the result is:
Ft 1 yt (1 ) y t 1 (1 ) 2 y t 2 (1 )3 y t 3 (1 )t 1 y1

Therefore, Ft+1 is the weighted moving

average of all past observations.
Simple Exponential Smoothing Method
The following table shows the weights assigned to
past observations for = 0.2, 0.4, 0.6, 0.8, 0.9
Weight assigned to 0.2 0.4 0.6 0.8 0.9
Yt 0.2 0.4 0.6 0.8 0.9

Yt-1 0.2(1-0.2) 0.4(1-0.4) 0.6(1-0.6)

2 2 2
Yt-2 0.2(1-0.2) 0.4(1-0.4) 0.6(1-0.6)
3 3 3
Yt-3 0.2(1-0.2) 0.4(1-0.4) 0.6(1-0.6)
4 4 4
Yt-4 0.2(1-0.2) 0.4(1-0.4) 0.6(1-0.6)
5 5 5
Yt-5 0.2(1-0.2) 0.4(1-0.4) 0.6(1-0.6)
Simple Exponential Smoothing Method
The exponential smoothing equation
rewritten in the following form elucidate the
role of weighting factor .
Ft 1 Ft ( yt Ft )

Exponential smoothing forecast is the old

forecast plus an adjustment for the error that
occurred in the last forecast.
Simple Exponential Smoothing Method
The value of smoothing constant must be
between 0 and 1
can not be equal to 0 or 1.
If stable predictions with smoothed random
variation is desired then a small value of is
desire.
If a rapid response to a real change in the pattern
of observations is desired, a large value of is
appropriate.
Simple Exponential Smoothing Method
To estimate , Forecasts are computed for
equal to .1, .2, .3, , .9 and the sum of
squared forecast error is computed for each.
The value of with the smallest RMSE is
chosen for use in producing the future
forecasts.
Simple Exponential Smoothing Method
To start the algorithm, we need F1 because
F2 y1 (1 ) F1
Since F1 is not known, we can
Set the first estimate equal to the first
observation.
Use the average of the first five or six
observations for the initial smoothed value.
Example:University of Michigan Index
of Consumer Sentiment
Date Observed
University of Michigan Jan-95
Feb-95
97.6
95.1
Index of Consumer Mar-95
Apr-95
90.3
92.5
Sentiment for May-95
Jun-95
89.8
92.7
January1995- Jul-95
Aug-95
94.4
96.2
December1996. Sep-95
Oct-95
88.9
90.2
Nov-95 88.2
we want to forecast the Dec-95 91
Jan-96 89.3
University of Michigan Feb-96
Mar-96
88.5
93.7
Index of Consumer Apr-96
May-96
92.7
94.7
Sentiment using Simple Jun-96
Jul-96
95.3
94.7
Exponential Smoothing Aug-96
Sep-96
95.3
94.7
Method. Oct-96
Nov-96
96.5
99.2
Dec-96 96.9
Jan-97
Example:University of Michigan Index
of Consumer Sentiment
Since no forecast is University of Michigan Index of Consumer
available for the first Sentiment

period, we will set the 100

Consumer Sentiment Index

first estimate equal to 98
96

the first observation. 94

92

We try =0.3, and 90

88

0.6. 86
Sep-94 Apr-95 Oct-95 May-96 Dec-96 Jun-97

Date
Example:University of Michigan Index
of Consumer Sentiment
Date Consumer Sentiment Alpha =0.3 Alpha =0.6
Jan-95 97.6 #N/A #N/A
Feb-95 95.1 97.60 97.60

Note the first forecast is Mar-95

Apr-95
May-95
90.3
92.5
89.8
96.85
94.89
94.17
96.10
92.62
92.55

the first observed value. Jun-95

Jul-95
Aug-95
92.7
94.4
96.2
92.86
92.81
93.29
90.90
91.98
93.43
Sep-95 88.9 94.16 95.09

The forecast for Feb. 95 (t Oct-95

Nov-95
90.2
88.2
92.58
91.87
91.38
90.67
Dec-95 91 90.77 89.19

= 2) and Mar. 95 (t = 3) Jan-96

Feb-96
Mar-96
89.3
88.5
93.7
90.84
90.38
89.81
90.28
89.69
88.98

are evaluated as follows: Apr-96

May-96
Jun-96
92.7
89.4
92.4
90.98
91.50
90.87
91.81
92.34
90.58
Jul-96 94.7 91.33 91.67

y t 1 y t ( yt y t ) Aug-96
Sep-96
95.3
94.7
92.34
93.23
93.49
94.58

y 2 y1 0.6( y1 y1 ) 97.6 0.6(97.6 97.6) 97.6

Oct-96 96.5 93.67 94.65
Nov-96 99.2 94.52 95.76
Dec-96 96.9 95.92 97.82

y 3 y 2 0.6( y2 y 2 ) 97.6 0.6(95.1 97.6) 96.1 Jan-97

Feb-97
97.4
99.7
96.22
96.57
97.27
97.35
Mar-97 100 97.51 98.76
Apr-97 101.4 98.26 99.50
May-97 103.2 99.20 100.64
Jun-97 104.5 100.40 102.18
Jul-97 107.1 101.63 103.57
Aug-97 104.4 103.27 105.69
Sep-97 106 103.61 104.92
Oct-97 105.6 104.33 105.57
Nov-97 107.2 104.71 105.59
Dec-97 102.1 105.46 106.55
 Example:University of Michigan Index
of Consumer Sentiment
RMSE =2.66 for = 0.6
RMSE =2.96 for = 0.3
University of M ichigan Index of Consumer sentiments

120

100

80
Sentiment Index

Consumer Sentiment
60 SES (Alpha =0.3)

SES(Alpha=0.6)

40

20

0
Jun-94 Oct-95 Mar-97 Jul-98 Dec-99 Apr-01

Months
Holts Exponential smoothing
Holts two parameter exponential
smoothing method is an extension of simple
exponential smoothing.
It adds a growth factor (or trend factor) to
the smoothing equation as a way of
adjusting for the trend.
Holts Exponential smoothing
Three equations and two smoothing
constants are used in the model.
The exponentially smoothed series or current level
estimate.
Lt yt (1 )( Lt 1 bt 1 )

The trend estimate.

bt ( Lt Lt 1 ) (1 )bt 1

Forecast m periods into the future.

F t m Lt mbt
Holts Exponential smoothing
Lt = Estimate of the level of the series at time t
= smoothing constant for the data.
yt = new observation or actual value of series in
period t.
= smoothing constant for trend estimate
bt = estimate of the slope of the series at time t
m = periods to be forecast into the future.
Holts Exponential smoothing
The weight and can be selected
subjectively or by minimizing a measure of
forecast error such as RMSE.
Large weights result in more rapid changes
in the component.
Small weights result in less rapid changes.
Holts Exponential smoothing
The initialization process for Holts linear
exponential smoothing requires two estimates:
One to get the first smoothed value for L1
The other to get the trend b1.
One alternative is to set L1 = y1 and
b1 y 2 y1
or
y 4 y1
b1
3
or
b1 0
Example:Quarterly sales of saws for
Acme tool company
Year Quarter t sales

The following table 1994 1

2
1
2
500
350
3 3 250

shows the sales of 1995

4
1
2
4
5
6
400
450
350

saws for the Acme 1996

3
4
1
7
8
9
200
300
350

tool Company. 2
3
4
10
11
12
200
150
400
1997 1 13 550
These are quarterly 2
3
14
15
350
250
4 16 550
sales From 1994 1998 1
2
17
18
550
400
3 19 350

through 2000. 1999

4
1
20
21
600
750
2 22 500
3 23 400
4 24 650
2000 1 25 850
2 26 600
3 27 450
4 28 700
Example:Quarterly sales of saws for
Acme tool company
Examination of the
plot shows: 900
Sales of saws for the Acme Tool Company: 1994-2000

A non-stationary time 800

700

series data. 600

Seasonal variation 500

Saws
400

seems to exist. 300

Sales for the first and 200

100

fourth quarter are larger 0

than other quarters. 0 5 10 15

Year
20 25 30
Example:Quarterly sales of saws for
Acme tool company
The plot of the Acme data shows that there might
be trending in the data therefore we will try Holts
model to produce forecasts.
We need two initial values
The first smoothed value for L1
The initial trend value b1.
We will use the first observation for the estimate
of the smoothed value L1, and the initial trend
value b1 = 0.
We will use = .3 and =.1.
Example:Quarterly sales of saws for
Acme tool company
Year Quarter t sales Lt bt Ft+m
1994 1 1 500 500.00 0.00 500.00
2 2 350 455.00 -4.50 500.00
3 3 250 390.35 -10.52 450.50
4 4 400 385.88 -9.91 379.84
1995 1 5 450 398.18 -7.69 375.97
2 6 350 378.34 -8.90 390.49
3 7 200 318.61 -13.99 369.44
4 8 300 303.23 -14.13 304.62
1996 1 9 350 307.38 -12.30 289.11
2 10 200 266.55 -15.15 295.08
3 11 150 220.98 -18.19 251.40
4 12 400 261.95 -12.28 202.79
1997 1 13 550 339.77 -3.27 249.67
2 14 350 340.55 -2.86 336.50
3 15 250 311.38 -5.49 337.69
4 16 550 379.12 1.83 305.89
1998 1 17 550 431.67 6.90 380.95
2 18 400 427.00 5.74 438.57
3 19 350 407.92 3.26 432.74
4 20 600 467.83 8.93 411.18
1999 1 21 750 558.73 17.12 476.75
2 22 500 553.10 14.85 575.85
3 23 400 517.56 9.81 567.94
4 24 650 564.16 13.49 527.37
2000 1 25 850 659.35 21.66 577.65
2 26 600 656.71 19.23 681.01
3 27 450 608.16 12.45 675.94
4 28 700 644.43 14.83 620.61
Example:Quarterly sales of saws for
Acme tool company
RMSE for this application
is: Quarterly Saw Sales Forecast Holt's Method

900

= .3 and = .1 800

700

RMSE = 155.5 600

The plot also showed the

500

Sales
sales
Ht+m
400

possibility of seasonal 300

200

variation that needs to be 100

investigated. 0
0 5 10 15

Quarters
20 25 30
Winters Exponential Smoothing
Winters exponential smoothing model is the
second extension of the basic Exponential
smoothing model.
It is used for data that exhibit both trend and
seasonality.
It is a three parameter model that is an extension
of Holts method.
An additional equation adjusts the model for the
seasonal component.
Winters Exponential Smoothing
The four equations necessary for Winters
multiplicative method are:
The exponentially smoothed series:
yt
Lt (1 )( Lt 1 bt 1 )
St s
The trend estimate:
bt ( Lt Lt 1 ) (1 )bt 1

The seasonality estimate:

yt
St (1 ) St s
Lt
Winters Exponential Smoothing
Forecast m period into the future:
Ft m ( Lt mbt )St ms
Lt = level of series.
= smoothing constant for the data.
yt = new observation or actual value in period t.
= smoothing constant for trend estimate.
bt = trend estimate.
= smoothing constant for seasonality estimate.
St =seasonal component estimate.
m = Number of periods in the forecast lead period.
s = length of seasonality (number of periods in the season)
Ft m = forecast for m periods into the future.
Winters Exponential Smoothing
As with Holts linear exponential smoothing, the
weights , , and can be selected subjectively or
by minimizing a measure of forecast error such as
RMSE.
As with all exponential smoothing methods, we
need initial values for the components to start the
algorithm.
To start the algorithm, the initial values for Lt, the
trend bt, and the indices St must be set.
Winters Exponential Smoothing
To determine initial estimates of the seasonal
indices we need to use at least one complete
season's data (i.e. s periods).Therefore,we
initialize trend and level at period s.
Initialize level as:
1
Ls ( y1 y2 ys )
s
Initialize trend as
1 y y y y2 y ys
bs ( s 1 1 s 2 ss )
s s s s
Initialize seasonal indices as:
y1 y y
S1 , S 2 2 ,, S s s
Ls Ls Ls
Winters Exponential Smoothing
We will apply Winters method to Acme
Tool company sales. The value for is .4,
the value for is .1, and the value for is
.3.
The smoothing constant smoothes the
data to eliminate randomness.
The smoothing constant smoothes the
trend in the data set.
Winters Exponential Smoothing
The smoothing constant smoothes the
seasonality in the data.
The initial values for the smoothed series Lt,
the trend bt, and the seasonal index St must
be set.
Example: Quarterly Sales of Saws for
Acme tool
Year Quarter t sales Lt bt St Ft+m
1994 1 1 500 1.333333
2 2 350 0.933333
3 3 250 0.666667
4 4 400 375 -12.5 1.066667
1995 1 5 450 396.9667 -9.05333 1.273412 483.3333
2 6 350 372.3747 -10.6072 0.935307 362.0524
3 7 200 296.7938 -17.1046 0.668827 241.1783
4 8 300 287.3869 -16.3348 1.059833 298.3352
1996 1 9 350 302.1219 -13.2278 1.23893 345.161
2 10 200 252.9623 -16.821 0.891905 270.2048
3 11 150 201.4173 -20.2934 0.691596 157.9377
4 12 400 268.2504 -11.5807 1.189227 191.9611
1997 1 13 550 373.5062 0.102908 1.309011 317.9958
2 14 350 363.8087 -0.87713 0.912946 333.2237
3 15 250 317.4823 -5.42206 0.720351 251.002
4 16 550 406.7605 4.047961 1.238103 371.1103
1998 1 17 550 465.9614 9.563264 1.270414 537.7528
2 18 400 444.9496 6.505758 0.908756 434.1286
3 19 350 410.5851 2.418728 0.759978 325.2062
4 20 600 487.3071 9.84905 1.236049 511.3412
1999 1 21 750 597.7855 19.91199 1.265679 631.5942
2 22 500 570.255 15.16774 0.899169 561.3363
3 23 400 510.9496 7.720431 0.766841 444.9085
4 24 650 570.7076 12.92419 1.206915 641.1016
2000 1 25 850 689.6728 23.52829 1.255716 738.6906
2 26 600 667.561 18.96428 0.899057 641.2886
3 27 450 591.6084 9.472591 0.764981 526.4561
4 28 700 640.1658 13.38107 1.172881 725.4539
Example: Quarterly Sales of Saws for
Acme tool
RMSE for this application Quarterly Saw Sales Forecas:t Winter's Method

is: 900

800

= 0.4, = 0.1, = 0.3 700

and RMSE = 83.36 600

500

Note the decrease in

Sales
sales

Ft+m
400

RMSE. 300

200

100

0
0 5 10 15 20 25 30

Quarters
Additive Seasonality
The seasonal component in Holt-Winters
method.
The basic equations for Holts Winters
additive method are:
Lt ( yt St s ) (1 )( Lt 1 bt 1 )
bt ( Lt Lt 1 ) (1 )bt 1
St ( yt Lt ) (1 ) St s
Ft m Lt bt m St m s
Additive Seasonality
The initial values for Ls and bs are identical
to those for the multiplicative method.
To initialize the seasonal indices we use
S1 y1 Ls , S2 y2 Ls ,, S s Ys Ls